Coming back after a break? Take a five-minute warm-up on Sections 1 to 4 before you start.
When an agent’s answer or finished work comes out wrong, these four questions help you find the cause:
- What was asked? Was your request clear, and did any instructions the agent followed, like a skill, say the right thing? Agents sometimes do a great job of solving a slightly different problem from the one you meant.
- What was it given? Were the files, data, and web pages it worked from the right ones? Check the dates, and look for anything out of date or cut off.
- What did it do? Did a tool fail partway while the agent carried on without telling you? Many apps let you open the list of steps an agent took, so you can scroll back through them and look.
- Where did it run? Did it work on your own computer or on its own computer in the cloud, and was it signed in to the accounts it needed? A file it saved in the cloud won’t be on your computer.
Each question points at a different part of the system from The Parts of an AI System:
